JP2000267853A - Method and device for retrieving examples - Google Patents

Abstract

PROBLEM TO BE SOLVED: To obtain a method and a device which can ideally retrieve a similar example that is difficult to be reached simply by tracing a diagnosis tree by retrieving similar sentences. SOLUTION: This method is provided with an example load step S1 where a cluster is loaded together with an example sentence associated with the cluster onto a memory from an example database 21, an example sentence analysis step 82 in which the example sentence and a retrieved sentence are analyzed, a processing request deciding step S8 in which requests are decided to the example sentence and the retrieved sentence on the basis of an input from a user interface and the retrieval processing of tree structure retrieval or similar sentence retrieval is performed, a sentence similarity calculating step S3 in which the similarity of the retrieved example sentence is calculated to the retrieved sentence, a cluster similarity calculating step S4 in which similarity in each cluster is calculated on the basis of the similarity calculated in the step S3 and a display information updating step S12 in which retrieved results are shown on a user interface.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an electronic case text stored on a computer, in particular, to store various inquiries and replies from a customer such as a help desk service, and to a failure diagnosis. More particularly, the present invention relates to a case search method and a case search device for searching using a result of classifying cases in which a history corresponding to a failure in a specific task is accumulated.

[0002]

2. Description of the Related Art A search method based on decision tree creation is a typical example of a conventional case search method. A case is represented by a plurality of attributes and their attribute values, and is characterized in that a set of cases is created by combining attributes and attribute values so as to minimize branching. However, this method
Since the relationship between the attribute and the attribute value is treated as an independent event, as described in Japanese Patent Application Laid-Open No. Hei 9-146627, "Failure Diagnosis Knowledge Tree Hierarchy Apparatus," There was a problem that the relationship of what to do was not taken into account. On the other hand, Japanese Patent Application Laid-Open No. Hei 9-46627 discloses a method of searching for cases by creating a hierarchical diagnostic tree by inputting and accumulating cases as knowledge in a plurality of causal relationships, and organizing them. Suggests how to use it.

More specifically, at the time of accumulating the cases, the input is divided into causal relationships such as "signs", "actions", and "results". "Signs" and "Results"
Is presumed to be divided into attributes and attribute values and input, and "behavior" is presumed to be divided and input into items such as "what", "equipment / equipment", and "what to do". Then, the expert generates a diagnosis tree using ID3 which is a known algorithm while checking the hierarchy based on the temporal sequence of the causal relationship.

However, even in the above-mentioned method, if the "action" and "result" of the fault diagnosis are not performed according to a predetermined fixed procedure, an ideal diagnosis tree can be obtained even if the causal relationship of the fault diagnosis cases is hierarchized. It was difficult to create. Moreover, as a method of searching for a diagnosis tree at the time of failure diagnosis, only a method of tracing the hierarchy of the diagnosis tree is provided, and a similar case may not be reached unless an ideal diagnosis tree is created.

[0005]

As described above, in the conventional case retrieval method, generally, an attribute to be given to a case stored in advance is designated, and the attribute and the attribute value are determined based on the attribute and the attribute value. A decision tree is created using an algorithm such as ID3. And inquiries,
In the failure diagnosis work, since similar cases are found by tracing the decision tree, there is no guarantee that the contents to be confirmed as the inquiry work or the failure diagnosis work can be used as a decision tree in an appropriate order. Further, Japanese Patent Application Laid-Open No. 9-146627 proposes classifying cases by focusing on a time axis as a causal relationship of the cases and using the classified cases for fault diagnosis work. It is assumed that the input is a simple template. However, in actual work, it was difficult to fill a complex template.

The present invention has been made in consideration of the above circumstances, and is used as an inquiry business history and a failure diagnosis business history while tracing a case database constructed as a diagnosis tree from case data accumulated daily. But not
By tracing the diagnostic tree by searching for similar sentences,
It is an object of the present invention to obtain a case search method and a case search device that enable a suitable case to be searched and found for a similar case that has been difficult to reach.

[0007]

In a case search method according to the present invention, a case of an inquiry history or a failure diagnosis history is stored in a case database using a case database that has been clustered in advance and stored. A case loading step of loading a cluster together with a cluster structure and case sentences associated with each cluster onto a memory; analyzing the case sentence loaded in the case loading step; and analyzing a search sentence input at the time of case search. A processing request that determines a request based on an input from a user interface for a case sentence analysis step and a case sentence and a search sentence analyzed by the case sentence analysis step, and performs a search process of a tree structure search or a similar sentence search. The request in the judgment step and the processing request judgment step A sentence similarity calculation step for calculating the similarity of the searched case sentence with respect to the search sentence, and a similarity between each cluster based on the similarity calculated in the sentence similarity calculation step. , And a display information updating step of displaying a search result on a user interface according to a request for a tree similarity search and a cluster similarity calculation step.

A similar cluster extraction step for extracting a cluster similar to each cluster based on the similarity obtained in the cluster similarity calculation step, and a case where the request in the processing request determination step is a similar cluster path display. And a similar cluster path calculating step of calculating a similarity degree of a cluster path based on information of the similar cluster extracted by the similar cluster extracting step.

Further, a conflicting cluster extraction step for extracting conflicting clusters that match a conflicting condition defined in advance in a thesaurus, and a conflicting cluster extraction for extracting conflicting clusters when the request in the processing request determination step is a conflicting cluster display The method further includes steps.

[0010] Further, a search history holding step for holding the items confirmed by the operator and the history additionally edited by the operator is further provided.

In the case retrieval apparatus according to the present invention,
In a case search device that uses a case database in which cases of inquiry histories and failure diagnosis histories are clustered in advance and stored, a cluster is stored in a memory together with a cluster structure and a case sentence associated with each cluster from the case database. A case loading means to be loaded, a case sentence analyzed by the case sentence analyzing means, a case sentence analyzing means for analyzing the case sentence loaded by the case loading means, and a search sentence inputted at the time of case search. In response to this, a request is determined based on an input from the user interface, and a search request for tree structure search or similar sentence search is performed. A sentence similarity calculating means for calculating the similarity of the searched case sentence with respect to the search sentence, Cluster similarity calculating means for calculating the similarity between each cluster based on the similarity calculated by the similarity calculating means, and a display for displaying the search result on the user interface in accordance with the request of the cluster similarity calculating means and the tree structure search It has information updating means.

A similar cluster extracting unit for extracting a cluster similar to each cluster based on the similarity obtained by the cluster similarity calculating unit, and a case where the request by the processing request determining unit is a similar cluster path display. And a similar cluster extracting means for extracting a similar cluster together with a path between mutual clusters, and a similar cluster path calculating means for calculating a similarity of a cluster path based on information on the similar cluster extracted by the similar cluster extracting means.

Further, a conflict cluster extracting means for extracting conflict clusters which match a conflict condition defined in advance in a thesaurus, and a conflict cluster extraction for extracting conflict clusters when the request by the processing request determination means is a conflict cluster display Means are further provided.

Further, there is further provided search history holding means for holding items confirmed by the operator and histories additionally edited by the operator.

[0015]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiment 1 Hereinafter, embodiments of the present invention will be described with reference to the drawings. FIG. 1 is a configuration diagram of a case search device of the present invention. In FIG.
Is a case database that classifies and stores case examples of customer inquiries.
The result of clustering the product name and the contents of the response to the sentence in the case in chronological order is accumulated. 22
Is a word dictionary, which is used as an analysis dictionary in morphological analysis and syntax analysis processing. Reference numeral 23 denotes a thesaurus describing the relationship between the words, and detailed is_a, ha
A relationship such as s_a is described. The thesaurus 23 further describes conflicting conditions according to the task. 2
4 is a word dictionary 22 and a thesaurus 2 according to the business.
3 is an ontology that collectively expresses No. 3. Reference numeral 25 denotes a user interface that interactively advances various processes performed by an operator using the case search device in a search operation in response to an inquiry or a failure diagnosis operation.

Reference numeral 1 denotes case loading means for loading the structure of cases stored in the case database and the contents of each case into a memory. 8 is a user interface 25
Is a processing request determination means for determining the processing content designated by the operator and selecting each processing. Reference numeral 2 denotes a search request sentence when the processing request sent from the user interface 25 is determined to be a similar sentence search request by the processing request determination means 8, and a search request sentence stored in the memory by the case loading means 1. This is a case sentence analyzing means for analyzing a sentence and a representative sentence using the ontology 24. 3 is a sentence analysis result analyzed by the case sentence analysis means using the ontology 24;
This is a sentence similarity calculating means for calculating a similarity between a sentence analysis result of a sentence with a content tag in an already stored case. 4
Is a cluster similarity calculating means for calculating a cluster similarity based on the similarity calculated by the sentence similarity calculating means 3.

Reference numeral 5 denotes a similar cluster extracting means for extracting a cluster similar to each cluster based on the similarity of the representative sentence of the cluster calculated by the sentence similarity calculating means 4. 6 is
This is a conflicting cluster extracting unit that extracts a cluster that conflicts with each cluster based on a conflicting condition described in the ontology 24. Numeral 7 denotes an initializing means showing a series of initializing processes from the case loading means 1 to the reciprocal cluster extracting means 6. Reference numeral 9 denotes a similar cluster extracting unit that extracts the similar cluster extracted by the similar cluster extracting unit 5. 1
Reference numeral 0 denotes a similar cluster path calculation unit that calculates the similarity using the similar cluster extracted by the similar cluster extraction unit 5 as a path of the diagnostic tree.

Numeral 11 denotes a conflict cluster extracting means for extracting the conflict cluster extracted by the conflict cluster extracting means 6. Reference numeral 12 denotes a cluster similarity calculating unit 4, a similar cluster path calculating unit 10, a contradictory cluster extracting unit 11,
And a display information updating means for updating the information of the display tree structure when the search request is determined by tracing the diagnostic tree in the processing request determination processing 8. Reference numeral 13 denotes a search history holding unit that holds the content to be left as a search history among the processes performed as a result of the determination of the process content by the process request determination unit 8.

FIG. 2 is a flowchart in the case search method of the present invention. Each step corresponds to the means in the configuration diagram of FIG. 1, and reference numerals 1 to 13 in FIG. 1 correspond to reference numerals S1 to S13 in FIG.

FIG. 3 is a configuration diagram for constructing the case database 21 in the case search apparatus of FIG. In FIG. 3, the same or corresponding parts as those in the case search apparatus shown in FIG. 1 are denoted by the same reference numerals, and description thereof will be omitted. Reference numeral 31 denotes a cluster generation step of performing clustering based on the similarity calculated by the sentence similarity calculation means 3 to group cases, and setting a case sentence having the largest number of cases in a cluster as a representative sentence. Reference numeral 32 denotes a cluster / case sentence correcting unit that corrects the structure of the cluster, the order, correspondence, and contents of the case sentences in the case when the cluster generated by the cluster generating unit 31 is not appropriate. 33 is a case structure storage means for storing the case cluster structure constructed in the cluster generation means 31 and the cluster / case sentence editing step 32 in the case database 21. Reference numeral 26 denotes a construction interface for operating the cluster generating means 31 and the cluster / case sentence editing means 32 when constructing the case database.

FIG. 4 is a flowchart for constructing the case database 21 in the case search method of the present invention. Each step corresponds to the means in the configuration diagram of FIG. 3, and 1 to 3, 8, 12, and 31 to 33 in FIG. 3 correspond to S1 to S3, S8, S12, and S31 to S33 in FIG.
It corresponds to.

Next, the operation will be described. FIG. 5 and FIG. 6 show examples of inquiries at the help desk of home electric appliances stored in the case database 21. FIG. 5 is an example of a case table for storing information for specifying a case. A case number 41 uniquely indicating each case, a product 42 indicating a product field of an inquiry, and a model name 4 indicating a specific product model.
3, a date 44 indicating the date of the inquiry and a title 45 briefly indicating the content of the inquiry are stored. FIG.
Is a case sentence table for storing information indicating the contents of inquiry and corresponding contents of each case in FIG. 5 for each sentence. A case number 46 for associating with the case in FIG. 5, the order of appearance of the sentences in the case , A content tag 48 indicating the content described in the sentence, and a tag sub-number 4 for setting the order of the content tag
9 and a case sentence 50 indicating a sentence of the actual recorded content. It is assumed that the case registrant corrects and registers the content tag given by the pattern matching from the surface expression of the case sentence at the time of storing the case, or the field for inputting the case is divided in advance for each content tag.

First, in order to enable the case retrieval apparatus of the present invention to perform processing, a case cluster is created by each step of FIG. The cases extracted from the case database 21 in the case loading step S1 are a set of cases that are simply grouped for each product and have no cluster in the initial stage. Clustering is performed on this case set in a case sentence analysis step S2, a sentence similarity calculation step S3, and a cluster generation step S31, using content tags determined in advance by the apparatus. For example, in the case of this embodiment, clustering is performed by calculating the similarity as a sentence for each tag sub-number using the content tag of "symptom" confirmed by the help desk operator to identify the cause of the inquiry from the customer. I do.

That is, the first-stage clustering is performed on the case sentences in which (content tag, tag sub-number) = (symptom, 1) of each case.
The second stage clustering is performed on the case sentence where (content tag, tag sub number) = (symptom, 2). At this time, the sentence similarity calculation step S3 is performed by the case sentence analysis step S2.
Dependency structure of each parsed sentence and thesaurus 2 containing information such as synonyms and related words contained in ontology 24
According to 3, the similarity is calculated based on the similarity of each word corresponding to the case structure.

For example, the similarity can be expressed by the following equation (1).

[0026]

(Equation 1)

Further, the clustering is performed by a known conventional method.

The result of the clustering performed in the cluster generation step S31 is displayed in a display information update step S12.
Is displayed on the construction interface 26. The case database builder checks the contents of each cluster,
If necessary, the cluster / case sentence editing step S32 performs processing such as merging and re-dividing clusters, changing the order of tag sub-numbers of case sentences, editing case sentences themselves, and editing case representative sentences. . When the case clusters are arranged, the case cluster structure is stored in the case database in a case structure storage step S33.

FIGS. 7 and 8 show examples in which the cases shown in FIGS. 5 and 6 are clustered and stored. FIG. 7 is an example of a cluster structure table, which includes a cluster number 61 uniquely indicating a cluster, a parent cluster number 62 indicating a parent cluster of the cluster, and a representative sentence 63 representing a cluster. The representative sentence is also used as a label for displaying the cluster in a tree structure. It is assumed that the representative sentence 63 has been modified at the time of construction in order to simply represent a cluster. In order to associate the cluster with the product, the product 6
4 is set for each cluster. Here, the value of the parent cluster 62 whose root is root indicates that the parent cluster 62 is a cluster of the first hierarchy. FIG. 8 is an example of a case statement table in a cluster, which is case number 6 corresponding to case number 41 in FIG.
6, a content tag 67 corresponding to the content tag 48 of FIG.
The tag sub-number 68 corresponding to the tag sub-number 49 of FIG.
Is the cluster number 69 corresponding to the cluster number 61 of FIG. In this way, a structured example of the stored cases is used at the time of case search.

FIG. 9 shows an example of a case structure stored in the case database 21 as a result of the case cluster construction. 7
Numeral 1 indicates a cluster of the first hierarchy, and indicates a cluster in which the column value of the cluster number 61 in FIG. 7 is 1.
Clusters 73 and 74 show examples of similar case clusters. Further, the clusters 75 and 76 are contradictory clusters, that is, an example in which when 75 is established, 76 is not established.

At the time of case search, the initialization step S in FIG.
A series of processing in step 7 is performed, and a diagnostic tree is configured on the memory. In the case loading step S1, first, the cluster structure table of FIG. 7 is loaded on a memory, and a hierarchical structure of a cluster is created from information of a product 60, a cluster number 61, and a parent cluster 62.

In FIG. 7, as one of the clusters of the first hierarchical level of the air conditioner, a cluster of cluster number 1 “the temperature of the air conditioner does not decrease” is generated, and its child cluster (second
As the (hierarchy), clusters of “No abnormal lamp is on” and “No abnormal lamp is blinking” of cluster numbers 2 and 3 are created.

Next, a set of (case number, content tag, tag sub number) is set to the cluster of the number stored in the cluster number 69 from the intra-cluster case statement table of FIG. In the cluster of cluster number 1 "air conditioner does not drop", (case number, content tag, tag sub number) =
(2, symptoms, 1), (5, symptoms, 1), (10, symptoms,
Set 1). Furthermore, from the case statement table of FIG.
The case sentence corresponding to (case number, content tag, tag sub number) is extracted from the column of the case sentence 50. For (case number, content tag, tag sub-number) = (2, symptom, 1), a case sentence "temperature does not decrease even if the set temperature of the air conditioner is lowered" is set.

Next, in a case sentence analyzing step S2, a representative sentence of each cluster and a case sentence set in each cluster are analyzed, and the analysis result is set on a memory. As the cluster of the cluster number 1, the analysis result of the representative sentence "the temperature of the air conditioner does not decrease" and the analysis result of the case sentence such as "the temperature does not decrease even if the set temperature of the air conditioner is decreased" are set.

In the sentence similarity calculation step S3, the similarity of all cluster representative sentences is obtained by comparing the analysis results of each cluster representative sentence. Furthermore, similar cluster extraction step S
In step 5, based on the similarity between the clusters, a cluster pair having a similarity equal to or greater than a certain threshold is defined as a similar cluster. When the processing of the similar cluster extraction step S5 is completed, in the conflict cluster extraction step S6, conflict clusters are extracted based on the conflict conditions described in the thesaurus 23 of the ontology 24.

FIG. 18 is an example of the contradictory condition described in the thesaurus 23. The contradictory conditions are registered by the case database builder when they are found at the time of constructing the case structure.
The “arrow” indicates the destination of the dependency structure of the sentence analysis result, and the left syntax structure 211 and the right syntax structure 212 are in conflict with each other. The cluster 73 and the cluster 74 in the example of FIG. 9 are similar clusters, and the cluster 75 and the cluster 76 are contradictory clusters from rule1 in FIG.

FIG. 10 shows clusters displayed at the time of case search. At the left end of the cluster, a confirmation check box 77 to be checked when the contents of the case statement are confirmed
Is displayed. In the center of the display, the number of cases classified into the cluster and the cluster content 78 of the representative sentence representing the cluster are displayed. If the cluster is the lowest level of the diagnostic tree, a lowest level cluster mark 79 is displayed at the right end.

FIG. 11 shows an example of a screen when starting a case search, a selection menu 81 for selecting a product, a text field 82 for performing a similar sentence search, and a similar sentence search button 83 for executing a similar sentence search. , A display switching button 84 for switching the display, a diagnostic tree display area 85 for displaying case clusters, and a history display button 86 for displaying a search history window. The display has two modes, “partial display” and “whole display”, and the display mode can be switched by the display switching button 84. “Partial display” displays only clusters that are intentionally expanded by the operator as shown in the screen of FIG. On the other hand, the “entire display” displays all the diagnostic trees shown in FIG.

When performing a case search, a product is selected on the selection menu 81 in accordance with the product of the inquiry. Here, it is assumed that a customer inquires about an air conditioner. The operator selects “air conditioner” in the selection menu 81 and listens to the contents of the inquiry from the customer.
In the case where the customer says "the temperature of the air conditioner does not reach the set temperature", a case cluster of "the temperature of the air conditioner does not drop" in the diagnostic tree display area 85 is selected. further,
Check the check box because the phenomenon of "air conditioner does not drop" has been confirmed. When these requests are received from the user interface 25, it is determined in the processing request determination step S8 that the displayed information should be corrected, and the displayed content is changed in the display information update step S12. Further, in the search history holding step S13,
In response to the check of the check box, the text “temperature of the air conditioner does not drop” is retained as the content tag “symptom”.

As a result, the screen is switched to the screen shown in FIG. 12, and the child cluster of the cluster 91 is displayed. Therefore, the operator understands that it is sufficient to check with the customer whether or not the abnormal lamp is on. If the customer answers that the abnormal lamp is not on, the operator selects the cluster 92 and selects the cluster 92. Click the check box. By repeating these operations, a diagnosis tree can be searched and a similar case can be reached.

The above is a case where a similar case can be retrieved only by tracing the diagnosis tree in response to an inquiry from a customer.
However, in some cases, the content of the customer's inquiry may not apply to the cluster of the first hierarchy in the diagnostic tree. For example, there is a case where the customer calls "anxious because the abnormal lamp is blinking". In such a case, the operator may enter text field 82
Is entered, and a similar sentence search button 83 is pressed. User interface 25
Then, when a similar sentence search is requested, in a processing request determination step S8, a case sentence analysis step S2 is called to analyze the search sentence. Further, in the sentence similarity calculation step, the similarity between all analyzed case sentences and sentences is calculated in the initialization step S7. Further, in a cluster similarity calculation step S4, a cluster similarity is calculated by the following equation (2) for a cluster whose similarity with the case sentence in the cluster exceeds a certain threshold.

[0042]

(Equation 2)

In this process, the degree of cluster similarity is increased with respect to a cluster which is considered to include many similar case sentences among clusters in which similar case sentences exist.

In the example of FIG. 9, the cluster 76 is searched as a cluster having a high similarity. The display information updating unit 12 switches to the “partial display” mode. Further, FIG.
In the user interface 25, only the diagnosis trees for the clusters 102 and 103 determined to be similar to the currently selected cluster 101 by the similar sentence search are displayed on the user interface 25, except for the cluster of the first hierarchy, as shown in the diagram on the right side of FIG. I do. At this time, by changing the color and displaying the color in the descending order of the similarity calculated in the cluster similarity calculation step S4, the operator can select the path of the diagnosis tree to be preferentially checked with the customer. Further, the operator can specify a similar case earlier by checking the contents of the cluster in the upper hierarchy in the diagnostic tree path selected preferentially. The operator switches to the “whole display” mode as necessary.

Further, it is conceivable that as a result of tracing the diagnostic tree, it is not possible to reach a related case. FIG.
The similar cluster path 105 is displayed from the display menu as shown by.
Is selected and displayed, a process for searching for a similar cluster path is called in a process request determination step S8. First, in a similar cluster extracting step, similar clusters of the cluster 111 and the cluster 112 are extracted. There is no similar cluster of the cluster 111, but two similar clusters of the cluster 112 are found. As a result, a path similar to the path to the currently selected case cluster is highlighted. At this time, the similarity of the similar cluster path is given by the following equation (3).

[0046]

(Equation 3)

In FIG. 15, the numbers in parentheses below the clusters in the right diagram are examples of S (i) in equation (3). The number in parentheses shown at the right end is Sk in equation (3). As a result, the operator can find a path of the diagnostic tree similar to the contents of the path that has been traced so far, and can find a similar case that could not be found by simply tracing the diagnostic tree.

For example, after the dialog between the operator and the customer,
In the diagnosis tree of FIG. 9, "the temperature of the air conditioner does not decrease", "the abnormal lamp is not lit", "the outdoor unit is receiving direct sunlight", "the temperature does not decrease even if the outdoor unit is shaded",
We arrived at the cluster at the end saying "there is enough space before the fan of the outdoor unit", but it did not solve even if we responded to the case of the cluster where "the space is enough before the fan of the outdoor unit" In this case, the cluster path 74 is regarded as a similar cluster path.
Is found and a new similar case is found.

Further, by selecting a conflicting cluster from the display menu of FIG. 14 and switching to displaying, the process is selected in the process determining step S8, and the cluster whose check box 77 is checked is selected in the conflicting cluster extracting step S11. Extract clusters that satisfy the extracted contradiction condition. In the example of the conflicting cluster display in FIG. 16, the conflicting cluster 122 and others are found with respect to the checked cluster 121, and the conflicting cluster is lightly displayed as shown in the right diagram. Clusters 75 and 7 in FIG.
In the case of No. 6, when the cluster 75 is confirmed, the color of the rejected cluster 76 is displayed lightly. This allows the operator to postpone the order of seeing the diagnostic tree branches that are unlikely.

Finally, after the operator has dealt with the customer using the diagnostic tree, the operator presses the history display button 86 to save the contents confirmed during the deal as a new case, as shown in FIG. Display the search history screen. 20
2 shows a representative sentence of a cluster checked as a search history together with a content tag. 202 is a list, which can be modified in the text field of 201 by selecting it. When editing is completed, click “Register”
By pressing the button 203, it is registered as a new case. With the list 202 selected, “operation” is selected from the operation menu, and processing such as “delete”, “insert”, and “move” can be performed. FIG. 17 illustrates a case 202 where the problem is solved by tracing the diagnostic tree of FIG. 9 from the cluster 71 to the cluster 73 as a result of the correspondence with the customer.

In this embodiment, in the initialization step S7
2, the similar cluster extracting step S5 and the contradictory cluster extracting step S6 were performed. However, the above processing is performed before the case structure storing step S33 shown in FIG. 4 and stored in the case database 21 to initialize FIG. It is also possible to load in step S7 and omit the similar cluster extraction step S5 and the contradiction cluster extraction step S6.

Further, the case sentence analyzed in the case sentence analyzing step S2 in FIG. 4 and the structure of the cluster representative sentence are stored in the case database 21 and loaded in the case loading step S1 in FIG. It is also possible to omit S2.

[0053]

According to the case search method according to the present invention, in the case search method using the case database in which the cases of the inquiry history and the failure diagnosis history are clustered in advance and stored, a cluster is created from the case database. A case loading step of loading on a memory together with a structure sentence associated with a structure and each cluster, and a case sentence analyzing step of analyzing the case sentence loaded in the case loading step and analyzing a search sentence input at the time of case search A processing request determining step of determining a request based on an input from a user interface for the case sentence and the search sentence analyzed by the case sentence analyzing step, and performing a search process of a tree structure search or a similar sentence search; If the request in the processing request judgment step is a request for similar case sentence search In addition, a sentence similarity calculation step for calculating the similarity of the searched case sentence with respect to the search sentence, and a cluster similarity for calculating the similarity between each cluster based on the similarity calculated in the sentence similarity calculation step A degree calculation step; a cluster similarity calculation step; and a display information updating step of displaying a search result on a user interface in accordance with a tree structure search request. For this reason, similar cases are also described for contents that the responder judged and assumed not to be the case when inputting the case, and for diagnosis trees that did not become definitive diagnosis trees because experts did not notice when constructing the case database. It is possible to search. In addition, like the inquiry from the customer, the top symptom of the diagnostic tree is not necessarily
Even when (sign) is not asked, there is an effect that it is possible to reach a similar case by searching for a similar symptom. Further, the search history is held for the contents checked and checked on the diagnosis tree, and the operator can display and edit the search history, so that the search operations performed can be easily summarized.

A similar cluster extraction step for extracting clusters similar to each cluster based on the similarity calculated in the cluster similarity calculation step, and a case where the request in the processing request determination step is a similar cluster path display And a similar cluster path calculating step of calculating a similarity degree of a cluster path based on information of the similar cluster extracted by the similar cluster extracting step. Therefore, a similar case can be searched for a diagnostic tree that has not become a definitive diagnostic tree.

A conflicting cluster extraction step for extracting conflicting clusters that match a conflicting condition defined in advance in a thesaurus, and a conflicting cluster extraction for extracting conflicting clusters when the request in the processing request determination step is a conflicting cluster display The method further includes steps. Therefore, it is possible to search for similar cases even for conflicting cases.

Further, there is further provided a search history holding step of holding an item confirmed by the operator and a history additionally edited by the operator. Therefore, the items confirmed by the operator and the history added and edited by the operator can be registered as new cases.

In the case retrieval apparatus according to the present invention,
In a case search device that uses a case database in which cases of inquiry histories and failure diagnosis histories are clustered in advance and stored, a cluster is stored in a memory together with a cluster structure and a case sentence associated with each cluster from the case database. A case loading means to be loaded, a case sentence analyzed by the case sentence analyzing means, a case sentence analyzing means for analyzing the case sentence loaded by the case loading means, and a search sentence inputted at the time of case search. In response to this, a request is determined based on an input from the user interface, and a search request for tree structure search or similar sentence search is performed. A sentence similarity calculating means for calculating the similarity of the searched case sentence with respect to the search sentence, Cluster similarity calculating means for calculating the similarity between each cluster based on the similarity calculated by the similarity calculating means, and a display for displaying the search result on the user interface in accordance with the request of the cluster similarity calculating means and the tree structure search It has information updating means. For this reason, similar cases are also described for contents that the responder judged and assumed not to be the case when inputting the case, and for diagnosis trees that did not become definitive diagnosis trees because experts did not notice when constructing the case database. It is possible to search. In addition, like the inquiry from the customer, the top symptom of the diagnostic tree is not necessarily
Even when (sign) is not asked, there is an effect that it is possible to reach a similar case by searching for a similar symptom. Further, the search history is held for the contents checked and checked on the diagnosis tree, and the operator can display and edit the search history, so that the search operations performed can be easily summarized.

A similar cluster extracting means for extracting a cluster similar to each cluster based on the similarity obtained by the cluster similarity calculating means, and a case where the request by the processing request judging means is a similar cluster path display. And a similar cluster extracting means for extracting a similar cluster together with a path between mutual clusters, and a similar cluster path calculating means for calculating a similarity of a cluster path based on information on the similar cluster extracted by the similar cluster extracting means. Therefore, a similar case can be searched for a diagnostic tree that has not become a definitive diagnostic tree.

A conflicting cluster extracting means for extracting conflicting clusters which match a conflicting condition defined in advance in a thesaurus, and a conflicting cluster extracting means for extracting conflicting clusters when the request by the processing request judging means is a conflicting cluster display Means are further provided. Therefore, it is possible to search for similar cases even for conflicting cases.

Further, there is further provided search history holding means for holding items confirmed by the operator and histories additionally edited by the operator. Therefore, the items confirmed by the operator and the history added and edited by the operator can be registered as new cases.

[Brief description of the drawings]

FIG. 1 is a configuration diagram of a case search device of the present invention.

FIG. 2 is a flowchart illustrating a processing operation of a case search method according to the present invention.

FIG. 3 is a diagram showing a configuration of a case cluster structure creation of the present invention.

FIG. 4 is a flowchart showing a processing operation for creating a case cluster structure according to the present invention.

FIG. 5 is an example of a case table stored in a case database of the present invention.

FIG. 6 is an example of a case sentence table stored in the case database of the present invention.

FIG. 7 is an example of a cluster structure table stored in a case database of the present invention.

FIG. 8 is an example of an in-cluster case sentence table stored in the case database of the present invention.

FIG. 9 is an example of a diagnostic tree stored in the case database of the present invention.

FIG. 10 is an example of cluster display in the case search device of the present invention.

FIG. 11 is an example of an initial case search screen in the case search device of the present invention.

FIG. 12 is an example in which symptoms are confirmed by the case retrieval apparatus of the present invention and the following confirmation items are displayed.

FIG. 13 is an example conceptually showing a result of similar sentence search performed by the case search device of the present invention.

FIG. 14 is an example showing a method of selecting a display method in the case search device of the present invention.

FIG. 15 is an example conceptually showing a result of performing similar cluster path display by the case search device of the present invention.

FIG. 16 is an example conceptually showing a result of performing a conflict cluster display by the case search device of the present invention.

FIG. 17 is an example in which a search history is displayed after a search is performed by the case search device of the present invention.

FIG. 18 is an example of a contradictory condition in the case search device of the present invention.

[Explanation of symbols]

1 case loading means, 2 case sentence analyzing means, 3 sentence similarity calculating means, 4 cluster similarity calculating means, 5 similar cluster extracting means, 6 conflicting cluster extracting means, 8 processing request judging means, 9 similar cluster extracting means, 10
Similar cluster path calculating means, 11 contradictory cluster extracting means, 21 case database, S1 case loading step, S2 case sentence analyzing step, S3 sentence similarity calculating step, S4 cluster similarity calculating step, S5
Similar cluster extraction step, S6 conflict cluster extraction step, S8 processing request determination step, S9 similar cluster extraction step, 10 similar cluster path calculation step, 11 conflict cluster extraction step.

 ──────────────────────────────────────────────────の Continuing on the front page (72) Inventor, Yoshiyuki Aikawa 2-3-2 Marunouchi, Chiyoda-ku, Tokyo Inside Mitsubishi Electric Corporation (72) Inventor Yamahiko Ito 2-3-2, Marunouchi, Chiyoda-ku, Tokyo (72) Inventor Katsushi Suzuki 2-3-2 Marunouchi, Chiyoda-ku, Tokyo F-term (reference) 5B075 ND03 NK02 NK32 NK43 NR02 NR12 PP03 PP13 PP25 PQ02 PR06 QM08 UU06

Claims (8)

[Claims] 1. A case search method using a case database in which cases of an inquiry history and a failure diagnosis history are clustered in advance and stored, wherein a cluster is associated with a cluster structure and each cluster from the case database. A case loading step of loading on the memory together with the case sentence; a case sentence analyzing step of analyzing the case sentence loaded in the case loading step and analyzing a search sentence input at the time of case search; A request for determining a request based on an input from a user interface for the case sentence and the search sentence analyzed by the above, and performing a search process of a tree structure search or a similar sentence search; If the above request in step is a request for similar case sentence search A sentence similarity calculating step of calculating a similarity of the case sentence searched for the search sentence; and a cluster calculating a similarity with each cluster based on the similarity calculated in the sentence similarity calculating step. A case search method comprising: a similarity calculation step; and a display information update step of displaying a search result on a user interface in accordance with the cluster similarity calculation step and a tree structure search request. 2. A similar cluster extracting step of extracting a cluster similar to each cluster based on the similarity calculated in the cluster similarity calculating step, and the request in the processing request determining step is a similar cluster path display. A similar cluster extracting step of extracting a similar cluster together with a path between mutual clusters when there is a similar cluster path calculating step of calculating a similarity of a cluster path based on information of the similar cluster extracted in the similar cluster extracting step. The case search method according to claim 1, further comprising: 3. A conflicting cluster extracting step of extracting a conflicting cluster that matches a conflicting condition defined in advance in a thesaurus, and a conflicting cluster extracting step of extracting a conflicting cluster when the request in the processing request determination step is a conflicting cluster display. 3. The method according to claim 1, further comprising a cluster extracting step. 4. The case search method according to claim 1, further comprising a search history holding step of holding an item confirmed by the operator and a history additionally edited by the operator. 5. A case search apparatus using a case database in which cases of an inquiry history and a failure diagnosis history are clustered in advance and stored, wherein clusters are associated with the cluster structure and each cluster from the case database. Case loading means for loading on a memory together with a case sentence; case sentence analyzing means for analyzing the case sentence loaded by the case loading means and analyzing a search sentence inputted at the time of case search; Processing request determining means for determining a request based on an input from a user interface with respect to the case sentence and the search sentence analyzed by the above, and performing a search processing of a tree structure search or a similar sentence search; If the request by means is a request for similar case sentence search, search for the search sentence Sentence similarity calculating means for calculating the similarity of the above-mentioned sentence sentence; cluster similarity calculating means for calculating the similarity for each cluster based on the similarity calculated by the sentence similarity calculating means; A case search apparatus comprising: cluster similarity calculation means; and display information updating means for displaying a search result on a user interface in accordance with a tree structure search request. 6. A similar cluster extracting means for extracting a cluster similar to each cluster based on the similarity calculated by the cluster similarity calculating means, and the request by the processing request determining means is a similar cluster path display. Similar cluster extracting means for extracting a similar cluster together with a path between mutual clusters when there is a similar cluster path calculating means for calculating a similarity of a cluster path based on information of the similar cluster extracted by the similar cluster extracting means. The case search device according to claim 5, further comprising: 7. A conflicting cluster extracting means for extracting a conflicting cluster that matches a conflicting condition defined in advance in a thesaurus, and a conflicting cluster for extracting a conflicting cluster when the request by the processing request determination means is a conflicting cluster display. 7. The case retrieval apparatus according to claim 5, further comprising a cluster extracting unit. 8. The case search apparatus according to claim 5, further comprising search history holding means for holding an item confirmed by the operator and a history additionally edited by the operator.